python code should never crash on C level, thus the bug is most likely
in pycurl, or less likely in libcurl.

I suspect libcurl already removed this request from multi and crashes
if python tries to remove it again.

d.

On 13 June 2012 16:12, Zdenek Pavlas <zpavlas_at_redhat.com> wrote:
> Hi,
>
> libcurl does not dispatch signals to Python handlers when waiting
> for a reply from a dead server (dtto for a proxy or dns).
> I understand implementing this properly is not easy, so tried
> to work this around with CurlMulti():
>
> - curl.perform()
>
> + m = pycurl.CurlMulti()
> + m.add_handle(curl)
> + try:
> +   while True:
> +     if m.perform()[0] == -1: continue
> +     num, ok, err = m.info_read()
> +     if ok: break
> +     if err: raise pycurl.error(*err[0][1:])
> +     if m.select(0.1) == -1: raise KeyboardInterrupt
> + finally:
> +   m.remove_handle(self.curl_obj)
>
> This mostly works, but there are issues when curl blocks in select()
> and I hit ctrl-c.  m.remove_handle() then sometimes crashes or triggers
> this invalid free():
>
> *** glibc detected *** /usr/bin/python: free(): invalid pointer: 0x08c316b0 ***
> ======= Backtrace: =========
> /lib/libc.so.6[0x87bfb6]
> /usr/lib/libcurl.so.4(Curl_http_done+0x9b)[0xae60bb]
> /usr/lib/libcurl.so.4(Curl_done+0x106)[0xaf83b6]
> /usr/lib/libcurl.so.4(curl_multi_remove_handle+0xfb)[0xb091cb]
> /usr/lib/python2.7/site-packages/pycurl.so(+0x3d11)[0x337d11]
>
> I'm using python-pycurl-7.19.0-7.fc14.i686
>
> I can reproduce this almost consistently, but not in a simple setup.
> (it depends eg on how much printing was done, etc)
>
> Is this a bug in curl, or the above Python code is wrong?
>
> Thanks,
> Zdenek
> _______________________________________________
> http://cool.haxx.se/cgi-bin/mailman/listinfo/curl-and-python

_______________________________________________
http://cool.haxx.se/cgi-bin/mailman/listinfo/curl-and-python
Received on 2012-06-13